Container code examination is an essential step in the container flow management. To date, this step is mostly achieved by human visual inspections, which are however time-consuming and error-prone. We hence propose a new computer vision system for automated container code recognition. The proposed system consists of model construction and code recognition stages. In the model construction stage, we first incorporate a locally thresholding method with prior knowledge of code character geometry to segment the code characters, including English characters A-Z and numeric characters 0-9, from a training set of container images. With the segmentation results of each character, we subsequently construct its Eigen-feature model using the principal component analysis (PCA). In the recognition stage, the code characters are firstly segmented from the given container image. Each segmented character is then recognized by finding the best matched Eigen-feature model that maintains the minimal PCA reconstruction error of the character appearance. Experiments showed that the proposed method achieved the code recognition with a high recognition rate and little recognition time for each image automatically. Overall, our proposed system has great potential for improving the efficiency of container terminals as well as enhancing the container management.